In the past few days, there has been a blockbuster news in the technology industry, that is, OpenAI has launched the text generation tool Sora, which has caused a sensation around the world. How powerful is this tool, whether it is a revolution or an innovation, let me briefly talk about my thoughts on AI.
The SORA tool is different from the previous generation software, which many people may not understand, and it feels like there has been such a thing for a long time.
For example, the previous function of replacing the celebrity face ** and the beautification function are not generated by AI, and they are not intelligent. It is human beings who tell the computer how to do it through programming algorithms, no matter how vivid and realistic the result is, it is the human who tells how to do it, which is equivalent to a thousand steps, all of which are designed by humans in advance and are completely under control.
But SORA is not this logic, it only based on some text prompts provided by people, and then the AI algorithm understands these texts by itself, and then figurs out what you mean, and then generates ** that meets your needs as much as possible. Therefore, after giving a text prompt, you have no idea how AI generates **, and the result is unpredictable, and the AI model has human-like intelligence and can handle your needs independently.
So, is this cool and powerful Sora another AI revolution?
I don't think it's just an innovation, not a revolution. Because the underlying AI logic has not changed, the intelligence has not improved much. However, the large language model represented by GPT is a big breakthrough for the previous AI model, which can be regarded as a revolution. Originally, due to the limitation of computing power, most of the previous AI models were aimed at a fixed scene, image recognition, speech recognition, weather**, the data and parameters involved were limited, and the functions were relatively targeted and relatively single. GPT large language model has benefited from the improvement of computing power in recent years, especially the wide application of GPU (which has created the legend of NVIDIA), which is to put a huge amount of data and parameters into it, black box training, and the AI model formed has the ability to understand human language, and intellectually speaking, it is closer to the human thinking mode, which is a revolution.
From chatgpt's text prompt text output, to **output, and then to the current **output, there is no major revolutionary upgrade in the underlying logic of the whole process, and it is just a different application of large language models.
However, with the continuous improvement of computing power and the deepening of AI training data, AI models will definitely become smarter than humans, and after crossing the critical point, it will become smarter than humans at an exponential rate, until it exceeds humans by a thousand times. The popularization of computers has greatly replaced human mental work and helped human beings save energy, but all output is predictable. If AI were much smarter than humans, the world would change dramatically. On the bright side, a lot of the work of deep human thinking is also done for others, including creative mental work; But the bad side is that the whole AI is no longer transparent to humans, and what it wants and does is gradually incomprehensible to humans, and the world at that time may become the world of AI, and whether or how humans exist or how will it exist will be a huge question mark.